Use the fake exception frame created by pmap_bootstrap2() by
capturing its return value and passing it along as the return
value from start_c_finish().
Move initialization of %dfc and %sfc into _bootstrap() (which will
eventually itself be re-factored to reduce duplication). Also, catch
up with the others: no need to initialize %usp before calling main().
For many many years now, it has been unnecessary to initialize the user
stack pointer before calling main(). TL;DR - %usp comes from the exception
frame and is restored in the syscall stub (always) or rei (if an AST is
pending). For newly-forked processes, those take the non-AST path through
rei, but lwp_trampoline() takes care of it in that case.
PR bin/60099 - Fix unquoted $* ($@) expansion
Make unquoted $* (or $@) when used in a context where field
splitting happens (which requires unquoted of course) generate
the same result, always, as would have been obtained had the
number of params been known, and
$1 $2 $3 ... ${n} (n is the value of $#)
been used instead of $* - so for x$*y, for any strings x & y
(including empty), the equivalent would be
x$1 $2 $3 ... ${n}y
This must hold, whatever the setting of IFS (which affects
how the fields generated are field split after they appear,
but should not affect the generation of those fields).
The bug this fixes is (in practice) never encountered, which is
how it has persisted for so long, and in so many different shells.
The method that was previously used to expand $* in this situation
would work perfectly, if the first character of ${IFS} was an IFS
[28 lines not shown]
Minor cleanups
Cleanup CTLccc char definitions (layout only):
#define [space|tab] consistency, align value column,
add hex equivalents (in comments) for octal char values
Minor changes to comments.
NFC.
Add DEBUG_ONLY() to include anything in DEBUG code
DEBUG_ONLY(x) expands to its arg, which should not contain commas
(outside parens) only in DEBUG mode compiles, most useful for var
declarations, and similar. Nb: code only, not outside functions.
Some cleanups/additions to comments.
By itself, this is a NFC change (until something uses DEBUG_ONLY()
for DEBUG mode code alteration).
PR lib/60111 - Add some bounds checking
As requested by the PR, bounds checking on the two obvious
places I could see where overflow looks possible has been added.
Detected overflow should generate a decode error.
PR 60119 - don't set -o pipefail (revert previous)
I was going to make the setting of the pipefail option check if
the executing shell supports the option or not (all should really,
and up to date dash does - it is a standard option) - but then I
looked at the script, and couldn't find a pipe used anywhere.
The option is useless here. So, bye-bye pipefail.
Reorder slightly for correctness
The "pipefail" option is now a POSIX standard option, move it
from the extensions section to the standard options section.
The (very fake) "r" option is not a standard option, though it
is a command line only "option" - move it away from -s, -c
the other (but standard) command line only options, and to a
section of its own (our first non-standard command line only option).
NFC - the order of options in this definition file is for human
(those who view this file) purposes only, all is reordered by
mkoptions.sh before being included with the shell sources.
Use sh internal char typing, not <ctype.h>
That is, use the internal shell is_digit() rather than isdigit().
No practical functional change (a digit is a digit).
Rather than pmap_boostrap2() returning the address of the lwp0 uarea,
let pmap_bootstrap2() construct the fake exception frame, initialize
lwp0.l_md.md_regs, and return the new lwp0 stack pointer, thus removing
a bunch of duplicated asm code in each platform's locore.s.
XXX amiga and atari TBD, but the changes will be less mechanical.
XXX sun2/sun3/sun3x TBD, require a slightly different set of mechanical
changes.
Update to version 9.2.219.
Changes:
- patch 9.2.0219: call stack can be corrupted
- patch 9.2.0218: visual selection highlighting in X11 GUI is wrong.
- patch 9.2.0217: filetype: cto files are not recognized
- patch 9.2.0216: MS-Windows: Rendering artifacts with DirectX
- patch 9.2.0215: MS-Windows: several tests fail in the Windows CUI.
- patch 9.2.0214: tests: Test_gui_system_term_scroll() is flaky
- patch 9.2.0213: Crash when using a partial or lambda as a clipboard provider
- patch 9.2.0212: MS-Windows: version packing may overflow
- translation(zh): Update the Simplify Chinese translation
- translation: align sponsor menu string spacing
- patch 9.2.0211: possible crash when setting 'winhighlight'
- runtime(fish): Add matchit support to filetype plugin
- patch 9.2.0210: tests: Test_xxd tests are failing
- runtime(doc): Update Italian xxd manpage
- patch 9.2.0209: freeze during wildmenu completion
- patch 9.2.0208: MS-Windows: excessive scroll-behaviour with go+=!
[49 lines not shown]
lowdown: update to 3.0.1.
Only 1 package depends on this.
Version 3.0.1, 2026-03-18
Flesh out -tmdoc to handle macros flush against other words and punctuation, e.g., (*-abcd*):.
Properly format -tmdoc and -tman compact lists, including nested compact lists.
